Celonis

What it is

Celonis is a process mining platform. It reads the event logs your systems already write — every order created, invoice blocked, price changed or approval granted in SAP, Oracle, Salesforce or ServiceNow — and reconstructs how a process actually runs, variant by variant, with the time spent between each step. The output is not the flowchart in the policy document; it is the dozens of paths an order really takes, which of them are slow, and which ones a person had to touch by hand.

It is not an automation tool, and it does not replace UiPath or n8n. It tells you where the time goes so you know what to automate first, and then measures whether the automation changed anything. Celonis is a Munich- and New York-based private company, last priced at nearly $13B when it raised $1B in August 2022. Gartner named it a Leader in its Magic Quadrant for Process Intelligence, published on 5 May 2026 and successor to the process mining report where Celonis had led for three years, positioned highest on ability to execute and furthest on completeness of vision among 13 vendors.

What changed for agents

Celonis now sells itself as the context layer that agents lack. An agent told to “fix late orders” knows nothing about which orders are late, why, or which fix worked last quarter; Celonis has that data. At Celosphere 2025 it launched a Process Intelligence MCP server, so an MCP client such as Claude can query process metrics and KPIs, search data models and write decisions back, rate-limited to 500 calls per minute per asset.

On 12 May 2026 it launched the Celonis Context Model, a live digital twin of operations built from process data and business knowledge, and signed an agreement to acquire Ikigai Labs, an MIT-linked forecasting and simulation company whose co-founder Devavrat Shah joins as chief scientist. The named agent platforms that can consume that context include Claude Cowork, Microsoft Copilot and Agent 365, Amazon Bedrock, IBM watsonx Orchestrate and Databricks Agent Bricks.

A use case worth buying it for

Order-to-cash across a CRM and an ERP. The quote closes in Salesforce, the order lands in SAP, and somewhere between them deals stall on credit blocks, pricing mismatches and manual re-keying. Celonis connects both event logs into one case per order, shows that, say, a third of orders pass through a manual price change, ties each variant to days-sales-outstanding, and flags the open orders at risk this week. That becomes the brief for the agent or bot you build next, and the baseline you measure it against. A Celonis-commissioned Forrester study (July 2025) credited its composite customer with raising the share of sales orders processed without human intervention from 33% to 86%; treat it as the vendor’s best case, not a forecast.

Pricing

No public price. Celonis’s own pricing link redirects to a sales contact form. The one public listing, on AWS Marketplace, meters “platform units” on 12-month contracts with up to 33% off for 36 months, and sizes the unit count by data volume and process scope in a private offer — it tells you the model, not the bill. Procurement benchmarks put an enterprise entry package around $150K–$250K a year before implementation, with implementation services quoted at a further $120K–$500K one-off. The same Forrester study reported $44.1M of benefits at 383% ROI over three years, which implies roughly $9M of three-year cost for its composite company. Budget in six figures to start, and expect it to grow with each process you add.

For comparison, Microsoft sells process mining inside Power Automate: Premium at $15 per user a month includes 50 MB of process-mining storage, and a $5,000-per-tenant-a-month add-on raises that to 100 GB.

Best for

Ops, finance-ops and transformation leaders at companies of 2,000 or more people running order-to-cash, procure-to-pay or service processes on SAP or Oracle, who want evidence of where the time goes before pointing agents or bots at a process — and who have a data engineer to own the connectors.

Alternatives, and when to pick them instead

• SAP Signavio — SAP’s own process intelligence suite, and a Leader in the same 2026 Magic Quadrant. Pick it when you are an SAP S/4HANA or RISE customer and want mining, modelling and SAP’s own data access in one contract.
• Microsoft Power Automate Process Mining — pick it when your processes live in Microsoft 365 and Dynamics, your data volume fits under 100 GB, and $60K a year is the ceiling.
• Pega Process Mining — also a 2026 Leader. Pick it when Pega already runs your case management and you want mining wired to the same workflows.
• UiPath — pick its process and task mining when the automation that follows will be UiPath bots anyway.
• Mimica — the funded newcomer, with a $26.2M Series B in September 2025, mining desktop activity rather than system logs. Pick it when the work happens in spreadsheets and email, where Celonis has no event log to read.
• Scribe — pick it when you need to document how one team does a task, not measure a cross-system process.

Watch-outs

• Your SAP data access is in litigation. Celonis is suing SAP in US federal court over the cost and technical barriers to extracting SAP data; SAP agreed in June 2025 not to interfere with the Celonis extractor or add fees until the case resolves, and trial is set for 7 December 2026. Guard: write data-access continuity and a price cap into your Celonis contract, and ask SAP in writing what extraction will cost you after the case.
• The platform is only as good as the connector. The first months go mostly to data engineering — mapping tables, cleaning timestamps, joining a CRM case to an ERP case. Guard: scope the first project to one process on one or two source systems, and get the connector running before you sign a multi-process expansion.
• Insight without an owner is shelfware. A dashboard showing dozens of order variants changes nothing if no one is accountable for cutting them. Guard: name a process owner and one target metric, such as touchless order rate or days-sales-outstanding, before go-live, and review it monthly.
• The Context Model is new. It launched in May 2026 and the Ikigai forecasting capabilities are still being folded in. Guard: buy on the process mining you can test today, and treat agent context as an upside you prove in a pilot, not a line in the business case.